One of the first studies about phonological awareness dimensionality was conducted by Stanovich, Cunningham, and Cramer (1984). In that study, 49 kindergarten children performed phonological awareness tasks differing in the unit of sound involved (rhyme and phoneme) and in their cognitive demand (production, detection, substitution, and deletion). The researchers performed an exploratory factor analysis considering the scores on the ten tasks administered and identified evidence of only one factor, which accounted for 48% of the variance in the data.
Wagner, Torgesen, Laughon, Simmons, and Rashotte’s (1993) study was one of the first to report evidence that phonological awareness may not be a one-dimensional construct. The study was conducted with 95 kindergartners and 89 second-graders. In the study, several phonological awareness tasks were administered, such as phoneme deletion, rhyme and alliteration detection, phoneme segmentation, blending onset and rhyme, blending phonemes into words, and blending phonemes into nonwords. Also, confirmatory factor analysis models were constructed to identify the model most consistent with the data. The results indicated that the two-correlated-factor model, with one factor indexed by three measures of blending (termed “phonological synthesis”) and another indexed by the other measures of phonological awareness (termed “phonological analysis”), demonstrated the best fit to the data.
Because of the inconsistency of the initial findings on the dimensionality of phonological awareness, several other studies on this topic were conducted, including, for example, the studies by Wagner and colleagues (Wagner et al., 1994; Wagner et al., 1997); Anthony and colleagues (Anthony et al., 2002; Anthony et al., 2011); Schatschneider et al. (1999); Hatcher and Hulme (1999); Runge and Watkins (2006); Papadopoulos and colleagues (Papadopoulos, Spanoudis, and Kendeou, 2009; Papadopoulos et al., 2012); and Vloedgraven and Verhoeven (2009). The results were mixed, with some studies corroborating the one-dimensionality of phonological awareness (Anthony et al., 2002, 2011; Papadopoulos et al., 2012; Schatschneider et al., 1999; Vloedgraven and Verhoeven, 2009; Wagner et al., 1997) and some studies corroborating the two-dimensionality of the construct (Hatcher and Hulme, 1999; Runge and Watkins, 2006; Wagner et al., 1994; Wolff and Gustafsson, 2015).
It should be noted that despite presenting consistent results, the studies that corroborated the one-dimensionality of phonological awareness (Anthony et al., 2002, 2011; Papadopoulos et al., 2012; Schatschneider et al., 1999; Vloedgraven and Verhoeven, 2009; Wagner et al., 1997) differed significantly regarding either the participants’ language, age, or the phonological awareness tasks administered. The languages included Dutch (Vloedgraven and Verhoeven, 2009), Greek (Papadopoulos et al., 2012), Spanish (Anthony et al., 2011), and English (Anthony et al., 2002; Schatschneider et al., 1999; Wagner et al., 1997). Considering age, for example, the study by Anthony et al. (2002) was conducted with younger children (2 and 3 years of age), and the study by Wagner et al. (1997) was conducted with older children who were already in the 4th grade. Most phonological awareness tasks included rhyme and phoneme awareness measures, with cognitive demand varying among the studies. Furthermore, and most importantly, the one-dimensional models were different across studies. For example, in Anthony et al. (2011), the best model was characterized by four factors related to the tasks’ cognitive demand (blending multiple-choice, blending free-response, elision multiple-choice, and elision free-response) loaded on a single, second-order factor. However, in Papadopoulos et al.’s (2012) study, the phonological awareness dimensionality was best captured by a nested-factor model, which consisted of a general first-order factor, a first-order supraphonemic factor, and a first-order phonemic factor, thus, taking into account the phonological unit involved in the tasks (e.g., rhyme, syllable, and phoneme).
On the other hand, the studies that reported evidence refuting the one-dimensionality of phonological awareness (Hatcher and Hulme, 1999; Runge and Watkins, 2006; Wagner et al., 1994) were also not consistent regarding the types of factors identified. For example, Wagner et al. (1994) replicated the findings of Wagner et al. (1993): two factors were identified, one termed “phonological synthesis” (indexed by the items of the measures of blending) and another termed “phonological analysis” (indexed by the items of the measures of phoneme deletion, rhyme, and alliteration detection and phoneme segmentation). In turn, in the study by Runge and Watkins (2006), although the researchers also identified two factors, the first factor included measures of phoneme detection and manipulation, and the second factor included measures of rhyme. Thus, whereas cognitive demand (analysis versus synthesis) defined the difference between the factors in the studies by Wagner et al. (1993, 1994), in Runge and Watkins, it was the phonological unit involved in the tasks (rhyme versus phoneme).
Hatcher and Hulme (1999) presented results more consistent with Runge and Watkins because, in their study, the measure of rhyme loaded on a factor different from the measures involving phoneme deletion, blending, and segmentation. Besides, recently, Wolff and Gustafsson (2015) employed phonological awareness tasks varying in cognitive demand (identification, blending/segmentation, and manipulation) and phonological unit (morpheme/word, syllable, and phoneme) and used confirmatory factor analysis to test different models about phonological awareness dimensionality. Although they refuted a one-dimensional model, their results corroborated a bifactor model with a general factor and four narrow factors representing the cognitive demand (blending/segmentation and manipulation) and the phonological unit (morphemes/words and phonemes) involved in the tasks.
Further complicating the issue of phonological awareness dimensionality, we shall consider that the degree of correspondence between graphemes and phonemes of a writing system can vary. This variation can affect the relationship between phonological awareness and reading (e.g., Ziegler et al., 2010). According to the psycholinguistic grain size theory (Ziegler and Goswami, 2005), since phonological awareness has a reciprocal relationship with reading, consistency in grapheme-phoneme mapping affects the development of phonological awareness (allowing the refinement of phonemic awareness). Consequently, it is possible that the dimensionality of phonological awareness could also be affected by the language’s writing system. Thus, research about the dimensionality of phonological awareness in understudied languages like Brazilian Portuguese is an important source of evidence for this debate.
According to Seymour, Aro, and Erskine (2003) classification, Portuguese has a simpler syllabic structure than Swedish and Dutch and is in an intermediate position concerning the consistency in the grapheme-phoneme mapping (between English, which has the less consistent grapheme-phoneme mapping, and Greek and Spanish, which have very consistent grapheme-phoneme mappings). Therefore, considering the languages in which phonological awareness dimensionality was studied so far (Greek—Papadopoulos et al., 2012, Spanish—Anthony et al., 2011, Swedish—Wolff and Gustafsson, 2015, Dutch—Vloedgraven and Verhoeven, 2009, and English—e.g., Anthony et al., 2002 and Hatcher and Hulme, 1999), Brazilian Portuguese adds information on variations in syllabic structure and consistency in grapheme-phoneme mapping.
Studies with Brazilian Portuguese
Regarding studies conducted with Brazilian Portuguese speakers, to our knowledge, three studies investigated the factorial structure of different measures of phonological awareness: Godoy and Cogo-Moreira (2015); Santos and Lima (2017); and Germano, César, and Capellini (2017). The study by Godoy and Cogo-Moreira (2015) included three measures of phonemic awareness (phoneme segmentation, onset deletion in nonwords with a consonant-vowel-consonant [CVC] structure, and onset deletion in nonwords with a consonant-consonant-vowel [CCV] structure); this study was conducted with 176 Brazilian 1st to 5th graders. The researchers assessed the fit of the data to a model in which the items of each of the three phonological awareness measures represented different factors that were correlated (segmentation, CVC deletion, and CCV deletion). The results indicated a good fit of this model to the data. The correlation among the three factors was strong. However, the authors did not compare the three-factor model with a one-factor model to assess which model best fits the data.
Santos and Lima (2017) aimed to investigate the evidence of construct validity of a phonological awareness instrument, the Roteiro de Avaliação da Consciência Fonológica (RACF). The RACF comprises three sections of items, each with five items to evaluate difficulties in detecting a phoneme at the beginning, middle, and end of words. Although the authors expected a single factor in the instrument, exploratory and confirmatory factor analysis suggested three factors in the RACF. These results are difficult to interpret theoretically because there is not a manipulation of phonological unit (e.g., syllable X phoneme) or cognitive demand (e.g., analysis X synthesis) in the RACF. Thus, as Santos and Lima acknowledge, the factors could be a byproduct of item difficulty (e.g., differential difficulty in detecting a phoneme at the beginning, end, and middle of the words).
Germano et al. (2017) evaluated the dimensionality of phonological awareness in the context of a screening protocol to identify children at risk of dyslexia. The study included several phonological awareness measures comprising rhyme production, rhyme identification, syllabic segmentation, production of words from a given phoneme, phonemic synthesis, and phonemic analysis. A total of 149 children aged 6 years to 6 years and 11 months were evaluated in phonological awareness and several other measures. An exploratory factor analysis was carried out, and four factors were retained. More importantly, the phonological awareness measures loaded mainly in two factors, one with measures of rhyme, alliteration and letter-naming knowledge, and the other with measures of phoneme analysis and synthesis, together with rapid automatized naming and word and nonword decoding. These results suggest that the phonological unit involved in the tasks (e.g., phonemic X supraphonemic) could be responsible for the two-dimensionality of phonological awareness. However, it is important to notice that syllabic segmentation did not load in the same factor as rhyme and alliteration. Besides, the inclusion of other variables not theoretically related to phonological awareness in the analysis makes it difficult to interpret the factors.
The present study
Considering the divergence among the results from studies on the dimensionality of phonological awareness conducted thus far in Brazil (Germano et al., 2017; Godoy and Cogo-Moreira, 2015; Santos and Lima, 2017), and internationally (Anthony et al., 2002, 2011; Hatcher and Hulme, 1999; Papadopoulos et al., 2012; Runge and Watkins, 2006; Schatschneider et al., 1999; Vloedgraven and Verhoeven, 2009; Wagner et al., 1997; Wagner et al., 1994; Wolff and Gustafsson, 2015), we aim to investigate this issue. The present study included measures that differ in their cognitive demand (detection, blending, segmentation, and deletion) and the phonological unit (rhyme, syllable, and phoneme). Thus it allows assessing whether measures that differ only in their cognitive demand (e.g., blending compared with segmentation) or in the linguistic unit emphasized (e.g., syllable compared with phoneme) will be in the same factor or different factors. To investigate this issue, confirmatory factor analysis was employed to test five models. One model representing a strong one-dimensional hypothesis: a one-dimensional model with all measures indexing a single factor. Two oblique models representing weak one-dimensional hypotheses: a two-dimensional model with two correlated factors representing phonological units (phonemic X supraphonemic); and a two-dimensional model with two correlated factors representing cognitive demand (analysis X synthesis). Two orthogonal models tested strong two-dimensional hypotheses: a two-dimensional model with two uncorrelated factors representing phonological units (phonemic X supraphonemic); and a two-dimensional model with two uncorrelated factors representing cognitive demand (analysis X synthesis).
In addition, the present study also investigates longitudinally whether phonological awareness dimensionality would be the same in kindergarten and at the end of the first grade (formal reading instruction starts in Brazil in the first grade). Thus, it allows evaluating the effect of reading instruction on phonological awareness dimensionality. It is important to do this because, according to the psycholinguistic grain size theory (Ziegler and Goswami, 2005), phonological awareness has a reciprocal relationship with reading and phonemic awareness only develops along with the teaching of reading.
We have no knowledge of other studies that have conducted a comprehensive investigation of phonological awareness dimensionality in Brazilian Portuguese like this one. Theoretically, it is important to have data on the dimensionality of phonological awareness in different languages because this can clarify whether phonological awareness is a general metalinguistic ability that underlies the acquisition of literacy in any alphabetical language (Khalaf, Santi, Kulesz, Bunta, and Francis, 2019). From a more practical point of view, it is essential to evaluate the dimensionality of phonological awareness because it has implications for test building and psychological assessment in terms of the kind of items employed in phonological awareness’ tasks and their interpretation.